Pith. sign in

REVIEW 2 major objections 5 minor 1 cited by

DeMo: Decoupled Feature-Based Mixture of Experts for Multi-Modal Object Re-Identification

T0 review · 2 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read Multi-modal object re-identification improves by decoupling features and weighting them with attention-triggered experts.

desk verdict Solid empirical ReID paper with careful ablations; the decoupling mechanism is plausible but unverified, and the paper deserves a serious referee. read the letter →

arxiv 2412.10650 v1 pith:TFSJP5YM submitted 2024-12-14 cs.CV

classification cs.CV
keywords multi-modalobjectre-identificationmixtureofexpertsfeaturedecouplingattentiongatingRGB-NIR-TIRpersonvehiclevisiontransformer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes a three-part framework, DeMo, for identifying the same object across cameras when each image has RGB, near-infrared, and thermal-infrared views. It argues that previous fusion methods weaken modality-specific cues by letting shared information dominate, and that imaging quality changes dynamically across modalities. Its solution is to extract multi-granularity features, decouple them into unimodal-specific, bimodal-shared, and trimodal-shared streams, and then weight those streams with attention instead of a fixed gate. If the mechanism works as described, DeMo gives robust retrieval even when some modalities are absent, and it reports top results on three person and vehicle benchmarks.

What carries the argument

Three components carry the method. PIFE pools patch tokens and concatenates them with the class token of a vision transformer to get multi-granularity modality features. HDM uses seven learnable query tokens, one per modality or pair of modalities, with multi-head cross-attention over keys built from the corresponding tokens, producing seven decoupled features. ATMoE treats each decoupled feature as an expert, computes attention weights from a reduced query and stacked keys, splits each expert output into multiple heads, scales each chunk, and concatenates the results; the final feature concatenates all weighted experts and the PIFE features.

What would settle it

Measure pairwise cosine similarity or mutual information between the seven decoupled outputs on RGBNT201 after training; if the unimodal-specific and bimodal-shared streams remain highly correlated with each other, the claimed decoupling is not actually separating the information.

Watch

Extended reading notes

Core claim

DeMo claims that multi-modal object re-identification is best served not by fusing raw modality features but by first separating each modality's unique cues from cues shared with one or both other modalities, then letting a multi-head attention mechanism weight the seven resulting streams per instance. With a CLIP backbone, this reaches 79.0 mAP / 82.3 Rank-1 on RGBNT201, 86.2 / 97.6 on RGBNT100, and 49.2 / 59.8 on MSVR310, and remains competitive when one or more modalities are missing. The paper also reports that the gains come with only a modest parameter and FLOP increase over its baseline.

Load-bearing premise

The load-bearing premise is that the seven decoupled features are truly non-overlapping, because no loss in the paper enforces orthogonality or diversity among them.

Editorial extensions

If this is right

  • DeMo achieves the highest reported mAP and Rank-1 on RGBNT201, RGBNT100, and MSVR310 among the compared methods, with both ViT and CLIP backbones.
  • The framework stays competitive when one or two modalities are missing, suggesting it can tolerate sensor failure or occlusion.
  • The extra modules add less than 13MB of parameters and about 0.8G FLOPs over the baseline, indicating the improvement is not from brute-force capacity.
  • Dynamic attention weights vary across instances and heads, consistent with the claim that the model adapts to changing imaging quality.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the decoupling is real, then adding an explicit orthogonality or mutual-information penalty between the seven streams should push the reported numbers higher rather than degrade them; the paper currently does not train with such a penalty.
  • The same attention-triggered gating could be lifted to other multi-modal retrieval settings beyond RGB-NIR-TIR, including visible-infrared person ReID or multi-sensor vehicle identification with more than three modalities.
  • The claim that HDM prevents interference would be more directly tested by checking whether the unimodal-specific streams alone, without ATMoE, already outperform fused baselines; the paper's ablations combine HDM with downstream weighting.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes DeMo, a framework for multi-modal object Re-identification (RGB, NIR, TIR) that combines a Patch-Integrated Feature Extractor (PIFE), a Hierarchical Decoupling Module (HDM), and an Attention-Triggered Mixture of Experts (ATMoE). The authors report state-of-the-art results on RGBNT201 (79.0 mAP / 82.3 Rank-1), RGBNT100 (86.2 / 97.6), and MSVR310 (49.2 / 59.8) with a CLIP backbone, and they report consistent improvements over a baseline in ablations (Table 5), as well as robustness in missing-modality settings (Tables 3 and 4). The central claimed mechanism is that HDM decouples multi-modal features into non-overlapping modality-specific, bimodal-shared, and trimodal-shared streams, and that ATMoE adaptively weights these streams to handle dynamic imaging quality.

Significance. If the results hold, DeMo is a strong empirical contribution to multi-modal object ReID. The ablations are systematic and show that each proposed module (PIFE, HDM, ATMoE) contributes positively on RGBNT201, and the method achieves substantial gains over prior work, especially on the smaller MSVR310 dataset. The paper includes a parameter-efficiency analysis and a released codebase, which helps reproducibility. The main weakness is that the headline mechanistic claim of non-overlapping decoupling is not quantitatively verified; the empirical gains could plausibly arise from added capacity rather than genuine feature separation. The missing-modality experiments are also underspecified, making them difficult to reproduce or interpret fairly against methods with explicit reconstruction modules.

major comments (2)
  1. [Hierarchical Decoupling Module (Eqs. 3-8)] The paper's central claim that HDM produces non-overlapping decoupled features is not supported by quantitative evidence. The seven outputs are obtained by cross-attention over key sets that share tokens (e.g., K_RN and K_RNT both contain RGB and NIR tokens), and the overall loss in Eq. (17) contains no orthogonality, diversity, or mutual-information term to enforce separation. The only supporting evidence is Fig. 5, a qualitative activation-map example. The authors should either add a quantitative decoupling-quality evaluation (e.g., pairwise cosine similarity or centered kernel alignment among the D features on held-out data, or a comparison with an explicitly decoupled variant that includes a separation loss) or soften the language from 'non-overlapping' to 'hierarchically separated' throughout the abstract and introduction, since the empirical gains may partly arise from the extra capacity of seven cross-attention heads rather than from a genuine non-overlap property.
  2. [Multi-modal Object ReID with Missing Modalities (Tables 3, 4)] The missing-modality experiments are not reproducible because the manuscript never specifies the input representation for absent modalities. The model is trained on complete triples, but Tables 3 and 4 evaluate with one or two modalities missing; the text does not state whether missing inputs are zeroed, replaced by a learned mask token, or the corresponding tokens are dropped. Without this information, the comparison with TOP-ReID (which uses explicit reconstruction modules) is difficult to interpret. The authors should describe the exact input construction and the training/evaluation protocol for these settings.
minor comments (5)
  1. [Eq. (15)] In Eq. (15), the chunked expert outputs are written as E_R^h in R^{Np×c}, but the expert output E_R in Eq. (14) is a vector in R^C; the chunks should be in R^c (or the dimension accordingly clarified).
  2. [Eq. (17) and Table 5] The notation L_g([f_R, f_N, f_T]) is ambiguous: it is unclear whether the loss is applied to the concatenated vector of the three modality features or to each modality branch separately. Please clarify the training objective.
  3. [Throughout] Several typos and grammatical errors should be fixed, e.g., 'porposed' in Related Work, 'machanism' in Table 6 and the supplementary text, and 'Beseides' in supplementary Section B.
  4. [Figure 5 and Section 'Activation Maps of Decoupled Features'] The sentence 'D_RN highlights areas that differ from those in D_R and D_N, which are shared between IR and IN' is unclear; the intended meaning of 'IR' and 'IN' should be clarified.
  5. [Supplementary Figs. 7 and 8] The paper fixes the number of attention heads to H=4, but on RGBNT100 and MSVR310 the reported best configurations use H=32 and H=16 respectively. A brief justification for the fixed H=4 (e.g., robustness, consistency, or avoiding overfitting) would help readers assess the sensitivity of the main results.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central results are direct held-out benchmark evaluations, not fits disguised as predictions; self-citations and hyperparameter choices are not load-bearing.

full rationale

The paper's contribution is empirical: DeMo is trained on standard training splits and evaluated on held-out test splits of RGBNT201, RGBNT100, and MSVR310, with reported mAP/Rank-1 scores obtained at inference time. The architecture equations (1)-(17) define a feed-forward pipeline; no equation is defined in terms of the target metric, and no reported number is reconstructed from a fitted parameter by construction. Ablations in Tables 5, 8-11, and 14 compare adding or removing modules and alternative structures, which is standard empirical validation. The choice of 4 attention heads is made after ablations and then applied uniformly across datasets; this is benchmark tuning or a mild overfitting risk, not a circular derivation. The mechanistic claim that HDM produces 'non-overlapping' decoupled features is not enforced by an orthogonality or diversity loss, and the visual evidence in Figures 5 and 10 is indirect; however, this is a weakness in the strength of the mechanism claim, not a circularity, because the cross-attention construction does not define the claimed non-overlap as its own output. The paper cites prior works by overlapping authors, including TOP-ReID and Magic Tokens, but these citations are used as related work, comparison baselines, and motivation; they do not supply a uniqueness theorem or externally forced ansatz that makes the DeMo design circular. Missing-modality robustness in Tables 3-4 is also evaluated on test data. Overall, no load-bearing step reduces to its own inputs; the low score reflects only minor self-citation and hyperparameter selection concerns.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The method is an empirical deep learning system; its main defined parameters are learned from data. The ledger lists the hand-chosen design choices and the assumptions the mechanism story depends on. Learnable query tokens and attention weights are ordinary trainable parameters rather than independently testable entities, so no physical or conceptual entity is entered beyond the trained modules.

free parameters (3)
  • ATMoE multi-head count H = 4
    Selected after ablations on RGBNT201 (Table 6) and head sweeps on RGBNT100 and MSVR310 in the supplementary material; reported as best on RGBNT201 but not on the other datasets, where more heads score higher.
  • Training hyperparameters = learning rate 3.5e-4, encoder learning rate 5e-6, 50 epochs, batch sizes 64/128
    Chosen by hand without an ablation study; all reported benchmark numbers depend on them.
  • Pooling method in PIFE = Average pooling
    GeM pooling gives slightly better RGBNT201 numbers (79.1 versus 79.0 mAP, Table 7) but average pooling is chosen for simplicity, so this is a hand decision that slightly lowers the reported headline number.
assumptions (5)
  • ad hoc to paper Cross-attention with separately initialized learnable queries yields non-overlapping modality-specific, bimodal-shared, and trimodal-shared features without any explicit decoupling constraint.
    This is the core mechanism premise of HDM; the paper asserts it but never verifies it with a diversity or orthogonality metric.
  • domain assumption A single pretrained ViT or CLIP encoder can serve as the visual backbone for RGB, NIR, and TIR inputs after light fine-tuning.
    Used in PIFE, Eq. (1); no per-modality adaptation is described.
  • ad hoc to paper Attention weights computed from decoupled features reflect the momentary quality of each modality, so weighting by them improves retrieval.
    The ATMoE design and the dynamic weight visualization (Fig. 6) assume attention weights are quality proxies; no quality labels or direct correlation analysis support this.
  • ad hoc to paper The network can produce useful features when one or two input modalities are absent even though no missing-modality module is described.
    Missing-modality results in Tables 3 and 4 appear without specification of how absent inputs are handled in PIFE and HDM.
  • domain assumption Standard supervised re-identification losses, label-smoothing cross-entropy plus triplet loss, are sufficient to train the decoupling and gating modules.
    Objective Functions, Eqs. (16)-(17); no auxiliary loss for decoupling is used.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeMo: Decoupled Feature-Based Mixture of Experts for Multi-Modal Object Re-Identification." pith.science (2026). https://pith.science/paper/TFSJP5YM

@misc{pith2026241210650,
  author       = {Pith},
  title        = {Pith review of: DeMo: Decoupled Feature-Based Mixture of Experts for Multi-Modal Object Re-Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TFSJP5YM}},
  note         = {Machine review of arXiv:2412.10650}
}
read the original abstract

Multi-modal object Re-IDentification (ReID) aims to retrieve specific objects by combining complementary information from multiple modalities. Existing multi-modal object ReID methods primarily focus on the fusion of heterogeneous features. However, they often overlook the dynamic quality changes in multi-modal imaging. In addition, the shared information between different modalities can weaken modality-specific information. To address these issues, we propose a novel feature learning framework called DeMo for multi-modal object ReID, which adaptively balances decoupled features using a mixture of experts. To be specific, we first deploy a Patch-Integrated Feature Extractor (PIFE) to extract multi-granularity and multi-modal features. Then, we introduce a Hierarchical Decoupling Module (HDM) to decouple multi-modal features into non-overlapping forms, preserving the modality uniqueness and increasing the feature diversity. Finally, we propose an Attention-Triggered Mixture of Experts (ATMoE), which replaces traditional gating with dynamic attention weights derived from decoupled features. With these modules, our DeMo can generate more robust multi-modal features. Extensive experiments on three multi-modal object ReID benchmarks fully verify the effectiveness of our methods. The source code is available at https://github.com/924973292/DeMo.

Figures

Figures reproduced from arXiv: 2412.10650 by the authors.

Figure 2
Figure 2. The overall framework of our DeMo. We first employ a Patch-Integrated Feature Extractor (PIFE) to extract multi [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Detailed structure of ATMoE. Attention-Triggered Mixture of Experts To address the dynamic imaging quality and appropriately balance decoupled features across different instances, we introduce an Attention-Triggered Mixture of Experts (AT￾MoE). Unlike traditional MoE (Liu et al. 2024a) where weights are directly generated from decoupled features, we incorporate an attention mechanism. With attention-guided interacti… view at source ↗
Figure 4
Figure 4. Feature distributions with t-SNE (Van der Maaten and Hinton 2008). Different colors refer to different IDs. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (7 more)
Figure 6
Figure 6. Figure 6: Visualization of dynamic weights across instances. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Comparison with different heads on MSVR310. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 7
Figure 7. Figure 7: Comparison with different heads on RGBNT100. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 9
Figure 9. Figure 9: Dynamic weights in ATMoE on RGBNT100. Rank List Comparison with Different Modules. In [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Activation maps of decouple features in HDM. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Rank list comparison with different methods. (a) TOP-ReID. (b) Our proposed DeMo. [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Rank list comparison with different modules. (a) Baseline. (b) Baseline + PIFE. (c) Baseline + PIFE + HDM. (d) [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Modality Unified Attack for Omni-Modality Person Re-Identification

    cs.CV 2025-01 conditional novelty 6.0 of 10

    Modality-specific adversarial generators trained with metric disruption, simulated cross-modal, and collaborative multi-modal losses transfer to black-box single-, cross-, and multi-modality person re-id models, reach...

Reference graph

Works this paper leans on

65 extracted references · 16 canonical work pages · cited by 1 Pith paper

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    L.; Kiros, J

    Ba, J. L.; Kiros, J. R.; and Hinton, G. E. 2016. Layer normalization. arXiv preprint arXiv:1607.06450

  4. [4]

    Cai, W.; Jiang, J.; Wang, F.; Tang, J.; Kim, S.; and Huang, J. 2024. A Survey on Mixture of Experts. arXiv preprint arXiv:2407.06204

  5. [5]

    M.; and Xiang, T

    Chang, X.; Hospedales, T. M.; and Xiang, T. 2018. Multi-level factorisation net for person re-identification. In CVPR, 2109--2118

  6. [6]

    Chen, L.; Sun, R.; Yu, Y.; Du, Y.; and Zhang, X. 2024. Visible thermal person re-identification via multi-branch modality residual complementary learning. IVC, 105201

  7. [7]

    Chen, Y.; and Wang, L. 2024. eMoE-Tracker: Environmental MoE-based Transformer for Robust Event-guided Object Tracking. arXiv preprint arXiv:2406.20024

  8. [8]

    Chowdhury, M. N. R.; Zhang, S.; Wang, M.; Liu, S.; and Chen, P.-Y. 2023. Patch-level routing in mixture-of-experts is provably sample-efficient for convolutional neural networks. In ICML, 6074--6114

Show all 65 references
  1. [9]

    Crawford, J.; Yin, H.; McDermott, L.; and Cummings, D. 2023. UniCat: Crafting a Stronger Fusion Baseline for Multimodal Re-Identification. arXiv preprint arXiv:2310.18812

  2. [10]

    Dai, D.; Deng, C.; Zhao, C.; Xu, R.; Gao, H.; Chen, D.; Li, J.; Zeng, W.; Yu, X.; Wu, Y.; et al. 2024. Deepseekmoe: Towards ultimate expert specialization in mixture-of-experts language models. arXiv preprint arXiv:2401.06066

  3. [11]

    Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

  4. [12]

    Gui, Y.; Chen, M.; Su, Y.; Luo, G.; and Yang, Y. 2024. EEGMamba: Bidirectional State Space Models with Mixture of Experts for EEG Classification. arXiv preprint arXiv:2407.20254

  5. [13]

    Guo, J.; Zhang, X.; Liu, Z.; and Wang, Y. 2022. Generative and attentive fusion for multi-spectral vehicle re-identification. In ICSP, 1565--1572

  6. [14]

    He, Q.; Lu, Z.; Wang, Z.; and Hu, H. 2023. Graph-Based Progressive Fusion Network for Multi-Modality Vehicle Re-Identification. TITS, 1--17

  7. [15]

    He, S.; Luo, H.; Wang, P.; Wang, F.; Li, H.; and Jiang, W. 2021. Transreid: Transformer-based object re-identification. In ICCV, 15013--15022

  8. [16]

    Hendrycks, D.; and Gimpel, K. 2016. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415

  9. [17]

    Hermans, A.; Beyer, L.; and Leibe, B. 2017. In defense of the triplet loss for person re-identification. arXiv preprint arXiv:1703.07737

  10. [18]

    Hwang, C.; Cui, W.; Xiong, Y.; Yang, Z.; Liu, Z.; Hu, H.; Wang, Z.; Salas, R.; Jose, J.; Ram, P.; et al. 2023. Tutel: Adaptive mixture-of-experts at scale. MLS, 5: 269--287

  11. [19]

    Ioffe, S.; and Szegedy, C. 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In ICML, 448--456. pmlr

  12. [20]

    A.; Jordan, M

    Jacobs, R. A.; Jordan, M. I.; Nowlan, S. J.; and Hinton, G. E. 1991. Adaptive mixtures of local experts. Neural computation, 3(1): 79--87

  13. [21]

    Kuang, Z.; Zhang, H.; Cheng, L.; Liu, Y.; Huang, Y.; and Ding, X. 2024. Unity in Diversity: Multi-expert Knowledge Confrontation and Collaboration for Generalizable Vehicle Re-identification. arXiv preprint arXiv:2407.07351

  14. [22]

    Li, H.; Li, C.; Zhu, X.; Zheng, A.; and Luo, B. 2020. Multi-spectral vehicle re-identification: A challenge. In AAAI, volume 34, 11345--11353

  15. [23]

    Li, W.; Zhu, X.; and Gong, S. 2018. Harmonious attention network for person re-identification. In CVPR, 2285--2294

  16. [24]

    Li, X.; Li, Q.; Liang, F.; and Wang, W. 2023. Multi-granularity pseudo-label collaboration for unsupervised person re-identification. CVIU, 227: 103616

  17. [25]

    Li, Y.; Jiang, S.; Hu, B.; Wang, L.; Zhong, W.; Luo, W.; Ma, L.; and Zhang, M. 2024. Uni-MoE: Scaling Unified Multimodal LLMs with Mixture of Experts. arXiv preprint arXiv:2405.11273

  18. [26]

    Lin, B.; Tang, Z.; Ye, Y.; Cui, J.; Zhu, B.; Jin, P.; Zhang, J.; Ning, M.; and Yuan, L. 2024. Moe-llava: Mixture of experts for large vision-language models. arXiv preprint arXiv:2401.15947

  19. [27]

    Liu, T.; Liu, H.; Shang, F.; Yu, L.; Han, T.; and Wan, L. 2024 a . Completed Feature Disentanglement Learning for Multimodal MRIs Analysis. arXiv preprint arXiv:2407.04916

  20. [28]

    Liu, X.; Yu, C.; Zhang, P.; and Lu, H. 2023. Deeply coupled convolution--transformer with spatial--temporal complementary learning for video-based person re-identification. TNNLS

  21. [29]

    Liu, X.; Zhang, P.; Yu, C.; Lu, H.; and Yang, X. 2021. Watching you: Global-guided reciprocal learning for video-based person re-identification. In CVPR, 13334--13343

  22. [30]

    Liu, X.; Zhang, P.; Yu, C.; Qian, X.; Yang, X.; and Lu, H. 2024 b . A video is worth three views: Trigeminal transformers for video-based person re-identification. TITS

  23. [31]

    Lu, H.; Zou, X.; and Zhang, P. 2023. Learning progressive modality-shared transformers for effective visible-infrared person re-identification. In AAAI, volume 37, 1835--1843

  24. [32]

    Luo, H.; Gu, Y.; Liao, X.; Lai, S.; and Jiang, W. 2019. Bag of tricks and a strong baseline for deep person re-identification. In CVPRW, 1487--1495

  25. [33]

    Pan, W.; Huang, L.; Liang, J.; Hong, L.; and Zhu, J. 2023. Progressively Hybrid Transformer for Multi-Modal Vehicle Re-Identification. Sensors, 23(9): 4206

  26. [34]

    Qian, X.; Fu, Y.; Jiang, Y.-G.; Xiang, T.; and Xue, X. 2017. Multi-scale deep learning architectures for person re-identification. In ICCV, 5399--5408

  27. [35]

    Radenovi \'c , F.; Tolias, G.; and Chum, O. 2018. Fine-tuning CNN image retrieval with no human annotation. TPAMI, 41(7): 1655--1668

  28. [36]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual models from natural language supervision. In ICML, 8748--8763. PMLR

  29. [37]

    Rao, Y.; Chen, G.; Lu, J.; and Zhou, J. 2021. Counterfactual attention learning for fine-grained visual categorization and re-identification. In ICCV, 1025--1034

  30. [38]

    Shi, J.; Yin, X.; Chen, Y.; Zhang, Y.; Zhang, Z.; Xie, Y.; and Qu, Y. 2024 a . Multi-Memory Matching for Unsupervised Visible-Infrared Person Re-Identification. arXiv preprint arXiv:2401.06825

  31. [39]

    Shi, J.; Yin, X.; Zhang, Y.; Xie, Y.; Qu, Y.; et al. 2024 b . Learning commonality, divergence and variety for unsupervised visible-infrared person re-identification. In The Thirty-eighth Annual Conference on Neural Information Processing Systems

  32. [40]

    Shi, J.; Zhang, Y.; Yin, X.; Xie, Y.; Zhang, Z.; Fan, J.; Shi, Z.; and Qu, Y. 2023. Dual pseudo-labels interactive self-training for semi-supervised visible-infrared person re-identification. In ICCV, 11218--11228

  33. [41]

    Sun, Y.; Zheng, L.; Yang, Y.; Tian, Q.; and Wang, S. 2018. Beyond part models: Person retrieval with refined part pooling (and a strong convolutional baseline). In ECCV, 480--496

  34. [42]

    Szegedy, C.; Vanhoucke, V.; Ioffe, S.; Shlens, J.; and Wojna, Z. 2016. Rethinking the inception architecture for computer vision. In CVPR, 2818--2826

  35. [43]

    Van der Maaten, L.; and Hinton, G. 2008. Visualizing data using t-SNE. JMLR, 9(11)

  36. [44]

    N.; Kaiser, .; and Polosukhin, I

    Vaswani, A.; Shazeer, N.; Parmar, N.; Uszkoreit, J.; Jones, L.; Gomez, A. N.; Kaiser, .; and Polosukhin, I. 2017. Attention is all you need. NeurIPS, 30

  37. [45]

    Wang, Y.; Liu, X.; Zhang, P.; Lu, H.; Tu, Z.; and Lu, H. 2023. TOP-ReID: Multi-spectral Object Re-Identification with Token Permutation. arXiv preprint arXiv:2312.09612

  38. [46]

    Wang, Y.; Zhang, P.; Wang, D.; and Lu, H. 2024 a . Other tokens matter: Exploring global and local features of Vision Transformers for Object Re-Identification. CVIU, 244: 104030

  39. [47]

    Wang, Z.; Huang, H.; Zheng, A.; and He, R. 2024 b . Heterogeneous Test-Time Training for Multi-Modal Person Re-identification. In AAAI, volume 38, 5850--5858

  40. [48]

    Wang, Z.; Li, C.; Zheng, A.; He, R.; and Tang, J. 2022. Interact, embed, and enlarge: Boosting modality-specific representations for multi-modal person re-identification. In AAAI, volume 36, 2633--2641

  41. [49]

    Wei, S.; Luo, Y.; Wang, Y.; and Luo, C. 2024. Robust Multimodal Learning via Representation Decoupling. arXiv preprint arXiv:2407.04458

  42. [50]

    Wu, D.; Liu, Z.; Chen, Z.; Gan, S.; Tan, K.; Wan, Q.; and Wang, Y. 2025. LRMM: Low rank multi-scale multi-modal fusion for person re-identification based on RGB-NI-TI. ESWA, 263: 125716

  43. [51]

    Xu, B.; Liang, J.; He, L.; and Sun, Z. 2022. Mimic embedding via adaptive aggregation: Learning generalizable person re-identification. In ECCV, 372--388. Springer

  44. [52]

    Yang, B.; Chen, J.; and Ye, M. 2023. Towards Grand Unified Representation Learning for Unsupervised Visible-Infrared Person Re-Identification. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 11069--11079

  45. [53]

    Yang, B.; Chen, J.; and Ye, M. 2024. Shallow-Deep Collaborative Learning for Unsupervised Visible-Infrared Person Re-Identification. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 16870--16879

  46. [54]

    Ye, M.; Shen, J.; Lin, G.; Xiang, T.; Shao, L.; and Hoi, S. C. 2021. Deep learning for person re-identification: A survey and outlook. TPAMI, 44(6): 2872--2893

  47. [55]

    Yin, H.; Li, J.; Schiller, E.; McDermott, L.; and Cummings, D. 2023. GraFT: Gradual Fusion Transformer for Multimodal Re-Identification. arXiv preprint arXiv:2310.16856

  48. [56]

    Yu, C.; Liu, X.; Wang, Y.; Zhang, P.; and Lu, H. 2024 a . TF-CLIP: Learning text-free CLIP for video-based person re-identification. In AAAI, volume 38, 6764--6772

  49. [57]

    Yu, Z.; Huang, Z.; Hou, M.; Pei, J.; Yan, Y.; Liu, Y.; and Sun, D. 2024 b . Representation Selective Coupling via Token Sparsification for Multi-Spectral Object Re-Identification. TCSVT

  50. [58]

    Zhang, G.; Zhang, P.; Qi, J.; and Lu, H. 2021. Hat: Hierarchical aggregation transformers for person re-identification. In ACM MM, 516--525

  51. [59]

    Zhang, P.; Wang, Y.; Liu, Y.; Tu, Z.; and Lu, H. 2024 a . Magic tokens: Select diverse tokens for multi-modal object re-identification. In CVPR, 17117--17126

  52. [60]

    Zhang, Q.; Wei, Y.; Han, Z.; Fu, H.; Peng, X.; Deng, C.; Hu, Q.; Xu, C.; Wen, J.; Hu, D.; et al. 2024 b . Multimodal fusion on low-quality data: A comprehensive survey. arXiv preprint arXiv:2404.18947

  53. [61]

    Zheng, A.; He, Z.; Wang, Z.; Li, C.; and Tang, J. 2023. Dynamic Enhancement Network for Partial Multi-modality Person Re-identification. arXiv preprint arXiv:2305.15762

  54. [62]

    Zheng, A.; Wang, Z.; Chen, Z.; Li, C.; and Tang, J. 2021. Robust multi-modality person re-identification. In AAAI, volume 35, 3529--3537

  55. [63]

    Zheng, A.; Zhu, X.; Ma, Z.; Li, C.; Tang, J.; and Ma, J. 2022. Multi-spectral vehicle re-identification with cross-directional consistency network and a high-quality benchmark. arXiv preprint arXiv:2208.00632

  56. [64]

    Zhong, Z.; Zheng, L.; Kang, G.; Li, S.; and Yang, Y. 2020. Random erasing data augmentation. In AAAI, volume 34, 13001--13008

  57. [65]

    Zhou, K.; Yang, Y.; Cavallaro, A.; and Xiang, T. 2019. Omni-scale feature learning for person re-identification. In ICCV, 3702--3712

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.